Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Balanced Multi-Task Attention for Satellite Image Classification: A Systematic Approach to Achieving 97.23% Accuracy on EuroSAT Without Pre-Training

Aditya Vir

TL;DR

The paper tackles the challenge of achieving high EuroSAT accuracy without pre-training. It presents a three-stage architectural study culminating in a balanced multi-task attention mechanism that fuses Coordinate Attention (spatial) and Squeeze-Excitation (spectral) with a learnable fusion parameter $\alpha$, which converges to $0.57$. The resulting 12-layer model reaches $97.23$% test accuracy and $0.9692$ Cohen's Kappa on EuroSAT, approaching the performance of fine-tuned ResNet-50 while using no external data. Thorough ablations and confusion-pattern analyses demonstrate the value of balanced dual-path attention and progressive regularization, with public release of code and models to enable reproducibility and extension.

Abstract

This work presents a systematic investigation of custom convolutional neural network architectures for satellite land use classification, achieving 97.23% test accuracy on the EuroSAT dataset without reliance on pre-trained models. Through three progressive architectural iterations (baseline: 94.30%, CBAM-enhanced: 95.98%, and balanced multi-task attention: 97.23%) we identify and address specific failure modes in satellite imagery classification. Our principal contribution is a novel balanced multi-task attention mechanism that combines Coordinate Attention for spatial feature extraction with Squeeze-Excitation blocks for spectral feature extraction, unified through a learnable fusion parameter. Experimental results demonstrate that this learnable parameter autonomously converges to alpha approximately 0.57, indicating near-equal importance of spatial and spectral modalities for satellite imagery. We employ progressive DropBlock regularization (5-20% by network depth) and class-balanced loss weighting to address overfitting and confusion pattern imbalance. The final 12-layer architecture achieves Cohen's Kappa of 0.9692 with all classes exceeding 94.46% accuracy, demonstrating confidence calibration with a 24.25% gap between correct and incorrect predictions. Our approach achieves performance within 1.34% of fine-tuned ResNet-50 (98.57%) while requiring no external data, validating the efficacy of systematic architectural design for domain-specific applications. Complete code, trained models, and evaluation scripts are publicly available.

Balanced Multi-Task Attention for Satellite Image Classification: A Systematic Approach to Achieving 97.23% Accuracy on EuroSAT Without Pre-Training

TL;DR

The paper tackles the challenge of achieving high EuroSAT accuracy without pre-training. It presents a three-stage architectural study culminating in a balanced multi-task attention mechanism that fuses Coordinate Attention (spatial) and Squeeze-Excitation (spectral) with a learnable fusion parameter , which converges to . The resulting 12-layer model reaches % test accuracy and Cohen's Kappa on EuroSAT, approaching the performance of fine-tuned ResNet-50 while using no external data. Thorough ablations and confusion-pattern analyses demonstrate the value of balanced dual-path attention and progressive regularization, with public release of code and models to enable reproducibility and extension.

Abstract

This work presents a systematic investigation of custom convolutional neural network architectures for satellite land use classification, achieving 97.23% test accuracy on the EuroSAT dataset without reliance on pre-trained models. Through three progressive architectural iterations (baseline: 94.30%, CBAM-enhanced: 95.98%, and balanced multi-task attention: 97.23%) we identify and address specific failure modes in satellite imagery classification. Our principal contribution is a novel balanced multi-task attention mechanism that combines Coordinate Attention for spatial feature extraction with Squeeze-Excitation blocks for spectral feature extraction, unified through a learnable fusion parameter. Experimental results demonstrate that this learnable parameter autonomously converges to alpha approximately 0.57, indicating near-equal importance of spatial and spectral modalities for satellite imagery. We employ progressive DropBlock regularization (5-20% by network depth) and class-balanced loss weighting to address overfitting and confusion pattern imbalance. The final 12-layer architecture achieves Cohen's Kappa of 0.9692 with all classes exceeding 94.46% accuracy, demonstrating confidence calibration with a 24.25% gap between correct and incorrect predictions. Our approach achieves performance within 1.34% of fine-tuned ResNet-50 (98.57%) while requiring no external data, validating the efficacy of systematic architectural design for domain-specific applications. Complete code, trained models, and evaluation scripts are publicly available.
Paper Structure (35 sections, 6 equations, 2 figures, 2 tables)

This paper contains 35 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Research Questions
  4. Contributions
  5. Related Work
  6. Satellite Image Classification Benchmarks
  7. Attention Mechanisms in CNNs
  8. From-Scratch Training versus Transfer Learning
  9. Methodology
  10. Experimental Setup
  11. Baseline Architecture (Model 1)
  12. Design
  13. Training Configuration
  14. Results and Analysis
  15. CBAM-Enhanced Architecture (Model 2)
  16. ...and 20 more sections

Figures (2)

  • Figure 1: Representative samples from the EuroSAT RGB dataset showing all 10 land use and land cover classes. Each class contains approximately 2,700 images at 64$\times$64 pixel resolution from Sentinel-2 satellite imagery.
  • Figure 2: Evolution of confusion patterns across three architectural iterations. (a) Baseline model exhibits significant River-Highway confusion (27 total misclassifications). (b) CBAM-enhanced model resolves River-Highway confusion (8 errors) but introduces vegetation classification trade-offs. (c) Balanced multi-task attention model simultaneously addresses all confusion patterns, achieving 5 River-Highway errors with recovered vegetation classification accuracy.